I hit this warning fairly frequently when resuming 4.4-rc5:
ret = usermodehelper_read_trylock();
if (WARN_ON(ret)) {
dev_err(device, "firmware: %s will not be loaded\n",
name);
goto out;
}
It seems like it could be improved in two ways fairly easily:
1. Shouldn't firmware_class try the direct load before trying to get
the usermodehelper lock?
2. Why is the !NOWAIT case doing a trylock? That seems wrong.
--Andy
--
Andy Lutomirski
AMA Capital Management, LLC
On 12/20/2015 04:20 AM, Andy Lutomirski wrote:
> I hit this warning fairly frequently when resuming 4.4-rc5:
>
> ret = usermodehelper_read_trylock();
> if (WARN_ON(ret)) {
> dev_err(device, "firmware: %s will not be loaded\n",
> name);
> goto out;
> }
>
> It seems like it could be improved in two ways fairly easily:
>
> 1. Shouldn't firmware_class try the direct load before trying to get
> the usermodehelper lock?
>
> 2. Why is the !NOWAIT case doing a trylock? That seems wrong.
>
> --Andy
>
I saw the same warning coming from the bluetooth stack, see the thread
http://article.gmane.org/gmane.linux.kernel/1950940 and
http://article.gmane.org/gmane.linux.kernel/1940130.
At least in the bluetooth case, the conclusion was the firmware request
should not be happening at that point in time and the reset_resume function
should be implemented instead. I put that work aside and haven't picked
it up again.
Thanks,
Laura
On Mon, Dec 21, 2015 at 9:54 AM, Laura Abbott <[email protected]> wrote:
> On 12/20/2015 04:20 AM, Andy Lutomirski wrote:
>>
>> I hit this warning fairly frequently when resuming 4.4-rc5:
>>
>> ret = usermodehelper_read_trylock();
>> if (WARN_ON(ret)) {
>> dev_err(device, "firmware: %s will not be loaded\n",
>> name);
>> goto out;
>> }
>>
>> It seems like it could be improved in two ways fairly easily:
>>
>> 1. Shouldn't firmware_class try the direct load before trying to get
>> the usermodehelper lock?
>>
>> 2. Why is the !NOWAIT case doing a trylock? That seems wrong.
>>
>> --Andy
>>
>
> I saw the same warning coming from the bluetooth stack, see the thread
> http://article.gmane.org/gmane.linux.kernel/1950940 and
> http://article.gmane.org/gmane.linux.kernel/1940130.
> At least in the bluetooth case, the conclusion was the firmware request
> should not be happening at that point in time and the reset_resume function
> should be implemented instead. I put that work aside and haven't picked
> it up again.
That's exactly the issue I'm seeing, and I think you're right.
Still, I think the firmware_class code is wrong, too. If it wants to
warn, it should warn that firmware was requested too early in resume.
Using trylock as an excuse to fail a firmware load seems completely
bogus to me.
>
> Thanks,
> Laura
--
Andy Lutomirski
AMA Capital Management, LLC